TLDRs;
- Ant Group launched Ling-1T, an open-source AI model with one trillion parameters designed for advanced reasoning and code generation.
- Ling-1T scored 70.42% on the 2025 AIME, matching top-tier AI models in mathematical reasoning.
- The model uses a “mid-training + post-training” approach with over 20T tokens to boost logic and inference.
- Ant’s Ling AI family now spans Ling, Ring, Ming, and LLaDA-MoE, marking its deep commitment to open AI collaboration.
Ant Group has unveiled its most ambitious artificial intelligence system to date , Ling-1T, a trillion-parameter open-source large language model (LLM) that marks a major milestone in China’s rapidly advancing AI landscape.
Developed under Ant’s growing Ling AI model family, Ling-1T introduces new levels of performance in mathematical reasoning, code generation, and complex problem-solving, setting benchmarks that rival even the world’s most advanced AI systems.
The launch of Ling-1T reflects Ant Group’s ongoing vision to democratize artificial intelligence by making frontier models open and accessible to developers, researchers, and businesses worldwide. The company says the new model not only matches but in some cases exceeds existing open-source and proprietary models across key benchmarks.
Redefining Performance at Trillion Scale
Ling-1T’s trillion-parameter design places it among the largest open-source models ever released.
According to Ant Group, the system scored 70.42% on the 2025 American Invitational Mathematics Examination (AIME), a benchmark that tests deep reasoning and mathematical capability. This score puts Ling-1T on par with the best-in-class AI models globally, while maintaining efficient inference speed and cost performance.
Ant engineers achieved this breakthrough by adopting a “mid-training + post-training” pipeline, incorporating an advanced chain-of-thought mechanism that helps the model reason through problems step by step. Trained on over 20 trillion high-quality tokens, Ling-1T demonstrates an enhanced understanding of both natural and technical language, allowing it to tackle everything from coding tasks to scientific and educational content generation.
The company notes that the model performs especially well in logic-intensive and symbolic reasoning tasks, such as geometric optimization and algorithmic design, areas that often expose the limitations of traditional LLMs.
Open-Source Vision for Collaborative AI Growth
Ant Group’s open-sourcing of Ling-1T signals a commitment to collaborative AI, aiming to advance research and make powerful models accessible.
Anchored by Ling-1T and the earlier Ring-1T-preview, the Ling AI family spans non-thinking, reasoning, multimodal, and experimental models. CTO He Zhengyu emphasized that AGI is a public good, with open access enabling inclusive AI that benefits developers and society alike.
Building the Future of Inclusive AI
Beyond the technological feat, Ant Group’s latest development aligns with its broader mission to power the digital transformation of global industries. The company envisions AI as a universal utility that enhances access to information, automates complex reasoning, and enables smarter tools for consumers and small businesses alike.
With Ling-1T, Ant positions itself not just as a fintech giant but as a major player in global AI infrastructure, competing alongside OpenAI, Google DeepMind, and Anthropic. By blending massive computational power with open collaboration, Ant Group is laying the groundwork for a new era of intelligent, transparent, and accessible AI systems.
